<title>Comparison of neural network classifiers for optical character recognition</title>

Abstract

The recognition of handwritten characters is an important technology for document processing and for advanced user interfaces. Recent advances in Artificial Neural Network (ANN) classifiers have shown impressive pattern recognition results when using noisy data. One advantage of ANN algorithms is that they are parallel by design, which allows a natural implementation on high-speed parallel architectures.The availability of standard databases of handwritten characters permits a fair comparison between different OCR classifiers. This paper compares the classification performance of two popular ANN algorithms: Back Propagation and Learning Vector Quantization. A set of digits from the National Institute of Standards and Technology's Handwritten Database is used to test the two classifiers. Each algorithm's execution time and memory efficiency is also compared, based on an implementation for Adaptive Solutions' highly parallel CNAPS architecture. We also show that a fair comparison cannot be made between OCR research that does not use the same set of characters for testing. INTRODUCTIONHandwritten Optical Character Recognition is a problem that has attracted much interest by both academic and commercial researchers. Some commonly cited applications include handwritten insurance forms, bank checks, and tax forms. The difficulty of OCR is caused by the large variety of styles used by different people. The complexity of the problem is illustrated by the fact that a common indication of a person's identity is the shape of their handwriting. It is also noted that many people can not read their own handwriting, much less the writing of another individual. This paper describes research results on a subset of the domain of handwritten OCR. We evaluate the use of two popular artificial neural network classifiers, Back Propagation and Learning Vector Quantization (LVQ), for the use of isolated digit recognition. We first discuss some of the results of previous researchers. We then proceed by describing the ANN algorithms and the recognition results on a common data set. Finally we evaluate the ANN algorithms based on the memory requirements and training time while running on a SIMD parallel architecture. Previous workThere is a lot of activity in applying ANNs to Handwritten OCR[1][2][3]. One of the most important research results in the last few years is the findings by Martin and Pittman [3] on recognition of isolated digits by Back Propagation networks. Martin and Pittman found that one of the most significant factors that influence recognition accuracy is the size of the training set. They made an independent study in 08194-08 158/921$4.00 SPIE Vol. 1661 (1992) / 191 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 06/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx